SimNet Compiled FPGA Models

Description: A collection of 25 compiled neural network models (.xmodel format) for the Xilinx ZCU104 FPGA with a B4096 DPU accelerator (DPUCZDX8G ISA 2). The models are members of the SimNet family, which is a set of lightweight 1D convolutional neural networks (1D-CNNs) for real-time anomaly detection on sliding windows of rubber bushing time-series data. Models span five depth tiers (3–7 convolutional stages) and a parameter range from approximately 25 thousand to 510 million parameters. Models: Model Parameters Depth model-25k 25,385 3 model-39k 39,385 3 model-59k 58,713 3 model-85k 84,521 3 model-129k 128,545 3 model-198k 197,705 4 model-282k 282,169 4 model-440k 439,545 4 model-658k 657,921 4 model-973k 972,537 4 model-1_5m 1,473,537 5 model-2_2m 2,229,361 5 model-3_4m 3,378,601 5 model-5_1m 5,051,921 5 model-7_6m 7,583,625 5 model-11m 11,398,681 6 model-17m 17,283,305 6 model-26m 25,840,833 6 model-39m 38,802,553 6 model-58m 58,406,369 6 model-88m 88,211,025 7 model-132m 132,307,793 7 model-200m 199,565,105 7 model-boundary ~300,000,000 7 model-overflow ~510,000,000 PyTorch training checkpoints were quantised and compiled using pytorch-nndct 3.5.0 (Vitis AI 3.5.0, opset 17, INT8 post-training quantisation) targeting the ZCU104 B4096 DPU fingerprint. The .xmodel files require Vitis AI VART runtime >= 3.0 on aarch64 (dataset found here) All models accept a single inference window of shape (1, 512, 3), batch size 1, 512 time steps, 3 channels (x_meas, v_meas, F_meas) scaled by the provided StandardScaler. Output shape is (1,), a single anomaly score (sigmoid logit).